Scene-Graph Grounding and Editing on Noisy-Input Scene-Graph Grounding and Editing
82.2IoUQwen2.5-VL-7B (Ours)
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Qwen2.5-VL-7B (Ours)variant=Ours2026.03 | 82.2 | 98 | |
| Qwen2.5-VL-7B (CoT-SFT)variant=CoT-SFT2026.03 | 77.3 | 84.4 | |
| Qwen3-8B (Ours, for reference)variant=Ours, for reference2026.03 | 66.4 | 93.9 | |
| Qwen2.5-VL-7B (Ans-SFT)variant=Ans-SFT2026.03 | 61.9 | 66.7 | |
| Qwen3-VL-235B-Instructvariant=Instruct2026.03 | 42.8 | 67.9 | |
| Qwen3-VL-235B-Thinkingvariant=Thinking2026.03 | 37.8 | 70.1 |